What are between category discriminations?
Determining between objects
How is perception of objects different for humans and computers?
For humans it is perception of familiar items, for computers it is perception of familiar patterns
Does variations in retinal image lead to the apparent size and shape of an object changing?
No
What are the challenges for object recognition?
- Intra-class variations: there is lots of variation in a particular object (e.g. there are lots of different chairs and we can recognise them all as chairs)
- Recognising objects when only part of it is visible
- Recognising objects from different views
- Recognising objects from unusual views
What are template theories of 2D pattern matching?
- There is a mini copy or template in the long term memory of all known patterns
- We compare the stimuli to templates in memory for the one with the greatest overlap until a match is found
- when we see something new we try to match it to our templates
What are flaws with the template theory?
- are there too many templates?
- how do we decide what to store in long term memory?
- where do we stop?
- imperfect matches
- it cannot account for the flexibility of pattern recognition system
- comparison requires identical orientations, size, position of template to stimuli
What are prototype theories?
Modification of the template matching theory with flexible templates. It possesses the average of each individual characteristic to create a prototype for matching. No match is perfect. A criterion for matching is needed
What are the flaws with the prototype theory?
- how do we create it to begin with?
- how do we create the average?
- do we update the prototype as we go through life?
What is Franks & Bransford (1971) experiment to explain the prototype theory
They presented objects based on prototypes, the prototype was not shown. Yet participants were confident that they had seen the prototype. This suggests the existence of prototypes
What are feature theories of object recognition?
The pattern consists of a set of features or attributes. E.g. A = 2 straight lines and a connecting cross bar. But we also need to know about the relationship between the features
What are the flaws for the feature theory of object recognition?
When talking about more complex objects than letters it doesn’t really work?
What are structural descriptions?
They describe the nature of the components of a configuration and the structural arrangement of these parts. E.g. - Capital T = 2 parts, 1 horizontal, 1 vertical; vertical supports horizontal’ vertical bisects horizontal
How does 3D object recognition work?
- First you must interpret input to the visual system as coherent structures, segregated from one another and from the background
- this must be processed to give a structural description which can then be matched to the descriptions of visual objects stored in memory. If you get an adequate match then the objects feel recognisable to you
What is Marr’s computational approach an example of and what do we have to think about after?
It is an example of early image processing. We then have to think about how we match from what we see to what we have stored in memory
What is Marr & Nishihara’s theory of 3D object recognition?
Objects are comprised of cylinders and the structural description specifies relationships between cylinders. Each cylinder has an axis and the way in which others are joined are expressed as coordinates.
What are the flaws in Marr and Nishihara’s approach?
It doesn’t seem complex enough
What is Biederman’s theory of 3D object recognition?
Objects are composed of basic shapes called GEONS (geometrical ions) such as blocks, cylinders, arcs and wedges. There are approximately 36 different volumetric shapes. There are a small number of structural relationships between GENOS:
- relative size
- verticality
- centring
- reltive size of surfaces at joints
What is Biederman’s theory called?
The viewpoint invariant theory
According to Biederman what is helpful in segmenting visual images into parts?
Concave parts of an object’s contour (on the edge of an object)
What are the ‘non-accidental’ properties that Geons are specified in terms of?
- Curvature: points on a curve
- parallel: sets of points in parallel
- co-termination: edges terminating in a common point
- symmetry: versus asymmetry
- co-linearity: points in a straight line
What does the viewpoint invariant theory suggest object recognition does?
Use the relationship between GEONS to make up structural objects and compare it to familiar objects
What was Biederman’s experiment in support of his viewpoint invariant theory?
According to his model Biederman hypothesised that forms of degradation which disrupt the basis for identifying GEONS should make objects more difficult to recognise. So Biederman deleted edges at points where easily reinstated or difficult to determine. Participants were found to be slow and inaccurate at ‘non-recognisable’ but relatively good at ‘recognisable’.
According to the viewpoint invariant theory why does taking away points of concavity make an object more difficult to recognise?
It makes it more difficult to break down into constituent geons
Give examples of support for Biederman’s theory
- Vogels et al found some cortical neurons in monkeys that are sensitive to GEONS. They assessed the response of individual neurons in the inferior temporal cortex to changes in GEONS or changes in the size of an object. They found that some neurons responded more to GEON changes providing support for GEONS
